It is shown that policies within a bounded $\chi^2$ divergence from the proxy-feasible reference distribution admit an $N$-independent safety-hacking bound, and instantiate this general coverage-control principle with constrained pessimistic sampling.
Abstract
Inference-time pipelines often sample multiple outputs, filter them with a learned safety model, and return the proxy-feasible output with the highest learned reward. We show that this composition creates a two-stage failure: an imperfect safety proxy first contaminates the feasible set with unsafe outputs, and reward maximization can then amplify this residual contamination. We define \emph{safety hacking} as selecting an output that passes the learned constraint but violates the true safety criterion. For constrained Best-of-$N$ sampling, we derive finite-$N$ bounds governed by the joint upper reward tails of safe and unsafe outputs within the proxy-feasible set. If unsafe-but-feasible outputs have the heavier tail, safety hacking becomes asymptotically certain as $N$ grows, even when false-positive mass and average safety- and reward-proxy errors are arbitrarily small. We also show that policies within a bounded $\chi^2$ divergence from the proxy-feasible reference distribution admit an $N$-independent safety-hacking bound, and instantiate this general coverage-control principle with constrained pessimistic sampling. Coverage control limits amplification but cannot repair a contaminated feasible set: admitted unsafe outputs may still be favored, and regularized selection is not necessarily safer than constrained Best-of-$N$ for every reward proxy. Toy and language-model experiments characterize both contamination and its reward-tail amplification, which exposes an inherent difficulty in inference-time scaling with learned safety models.
Best-of-N (BoN) sampling is the simplest and most widely deployed inference-time alignment strategy, but it suffers from two distinct problems: reward hacking, in which the selected response exploits errors in the proxy reward model, and the absence of any privacy protection for the sensitive human preference data used to train that reward model. We show that a single intervention-adding calibrated noise to reward scores before selection-resolves both. Our first result, Private Best-of-N (PrivBoN), establishes that Gumbel noise at an appropriate scale simultaneously provides $\epsilon$-differential privacy and implements KL-regularized alignment. Whenever the privacy budget exceeds a critical threshold $\epsilon^*$, the privacy-mandated noise is the regret-optimal regularization, and privacy imposes zero additional alignment cost-matching the information-theoretic skyline of Huang et al. (2025). Because $\epsilon^*$ depends on an unknown coverage coefficient, we introduce Private Inference-Time Pessimism (PrivITP), which combines $\chi^2$-regularized rejection sampling with a two-phase Gaussian mechanism. PrivITP achieves ex-post $(\epsilon,\delta)$-DP with a privacy cost independent of the number of responses $n$, cleanly decouples the regularization parameter from the privacy parameter, and attains the skyline up to a noise-inflation term. Experiments across several language models, datasets, and reward models confirm our results: PrivBoN and PrivITP are scaling-monotonic (unlike BoN, which degrades past a critical $n$), and PrivITP matches or outperforms PrivBoN at equivalent privacy levels, with the largest gains in the strong-privacy regime.
Ishika Jain, Nandini Bhattad, Sayak Ray Chowdhury· 0 citations
We prove that Nakamoto proof-of-work consensus has a budget-determined population structure with three security regimes. We compose three layers: the imported Garay-Kiayias-Leonardos (GKL) sufficiency theorem [3]; new attack constructions, native to the GKL round model, showing that when adversarial hashrate weakly exceeds honest hashrate the private-chain attack violates Persistence with probability one and a censorship attack violates Liveness; and an economic model of $n$ heterogeneous miners under an operational-security threshold that determines the honest/adversarial partition endogenously. The composition yields a safe region where GKL guarantees apply, a margin region where they fail although honest miners retain a hashrate majority, and a majority-loss region where the certainty-style attacks apply. The block-rate side of the GKL condition is derived from difficulty adjustment, not assumed. Extensions cover probabilistic compromise, mining-pool aggregation with both safety-positive and safety-negative effects, a conditional adaptive-corruption upper bound (when adaptive reach contains the budget-determined static compromised set), and post-halving entry-exit dynamics.
C. S. Wright· International Conference on...· 0 citations
This paper introduces C LUE (Clustering and Experience-based Verification) , a training-free, non-parametric verifier that improves selection and reranking in Large Language Model outputs and finds that correct and incorrect solutions exhibit measurable geometric differences in their hidden-state trajectories.
Zhenwen Liang, Ruosen Li, Yujun Zhou et al.· Annual Meeting of the Associ...· 0 citations
LLM benchmark scores can be precise even when the observation protocol does not identify the behavioral property they are intended to measure. In a controlled, solver-grounded setting, we formalize a protocol-level identifiability audit over a finite behavioral policy class: given policies H, observation support O, and estimand $\tau$, we test whether O separates every pair with different $\tau$. The audit requires zero model calls and resolves our diagnostic case: base-only observation collapses seven frozen deterministic policies into one equivalence class; full support yields seven classes and no cross-estimand collisions; every leave-one-out support retains a constructive collision witness. Empirically, both constrained-generation variants have pair-validity 1.0, yet base accuracy and selective-response fidelity diverge - 0.620 versus 0.324 across six balanced oracle-transition directions (cluster-bootstrap 95% CI [0.600, 0.642] vs. [0.304, 0.345]) - and the gap recurs on a second deterministic source (0.646 vs. 0.331). The audit also synthesizes a minimum identifying support $O^*$ for the frozen policy class: two cells instead of the full 36-cell tensor. This case shows how evaluation-design validity can be checked structurally before model inference and why base correctness does not determine intervention-response fidelity.
Machine-learned predictions can speed up offline NP-hard optimization, but asking a predictor what to do amounts to asking it to solve the problem, and committing an unchecked prediction forfeits every worst-case guarantee. CASP (Certificate-Augmented Solution Pruning) instead asks which parts of the search space may be ignored, and accepts each answer only after a sound polynomial-time verifier has checked it, so correctness never depends on prediction quality. We develop the learning theory of this design. The verifier makes the induced loss class uniformly bounded, so certificate parameters are learnable from $\tilde O(\varepsilon^{-2}\log K)$ samples ($K$ the maximum instance size), whereas the unverified commitment class admits no distribution-free rate and, under cost spread $R$, none below $\Omega(R/\varepsilon^2)$. Filtering noisy predictions by verifiable confidence dominates the standard min-combiner, with a margin we compute in closed form, and the prediction stays useful even given the LP, because it breaks ties on degenerate optimal faces, where every symmetric LP policy, meaning one whose commitments depend on the instance only through the verifiable confidence values, provably stalls. Experiments on five problems test the theory's quantitative predictions. With trained predictors, unverified pruning loses up to $26%$ of the optimum under distribution shift, while the verified deployment of the same predictions loses nothing.
Recent research on Large Language Model (LLM) safety has widely adopted guardrails to identify unsafe LLM outputs. Existing guardrails typically formulate safety assessment as a deterministic classification task, mapping a discrete token sequence to a discrete safety label. However, this paradigm has two limitations: First, safety assessment is inherently an uncertain problem, particularly during the early generation state. Second, relying solely on discrete token sequences discards the rich probabilistic information embedded in the LLM output distribution. To address these limitations, we propose the first completely probabilistic architecture-agnostic guardrail \textsc{ProbGuard} to leverage the LLM early output distributional signals for estimating and calibrating the safety probability, thereby enabling early stopping of unsafe ongoing outputs. Specifically, given an LLM's generated prefix distribution, we formulate the safety risk as the unsafe probability of its continued generation dynamics and estimate this risk by Monte-Carlo sampling. Through post-training on the distributional signals and calibrated safety risk, \textsc{ProbGuard} achieves the best calibration performance across all nine model--dataset combination settings, reducing the average Brier score and ECE by 79.6\% and 71.9\%, respectively, over the best baseline. \textsc{ProbGuard} further limits the attack success rate to at most 1\% across six representative jailbreak attacks after observing the LLM early output distributions from only the first ten decoding steps.
Xinzhe Huang, Biwu Yao, Kedong Xiu et al.· 0 citations
What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduAug 31, 2026
With millions of users across the world, Julia has been used to conduct cutting-edge research and to design new drugs, jet engines, heat pumps, and more.